Security and privacy
Encryption
Which secrets Nullgap encrypts and how, which values are stored only as hashes, and how data is protected in transit.
Nullgap has to keep some secrets in a form it can use: the token it calls your Jira with, the signing secret of your webhook. Those are encrypted before they are stored. Other secrets only ever need to be recognised, not read back: those are stored as hashes. This page lists both, says how the encryption works, and is plain about what it does and does not protect.
#Secrets that are encrypted
| Secret | Belongs to |
|---|---|
| Integration credentials: the Slack webhook address, the Jira API token, the Linear API key | Organisation |
| The secret used to verify events coming in from Jira | Organisation |
| Code-host credentials you entered: access tokens and webhook secrets for GitLab, Bitbucket Cloud and Azure DevOps, and the app private key and webhook secret for GitHub Enterprise Server | Organisation |
| Signing secrets of outbound webhooks | Organisation |
| The signing secret of the Slack assistant | Organisation |
| Test accounts you give for live testing | Organisation |
| The single sign-on (OIDC) client secret | Organisation |
| Authenticator app (TOTP) secrets | The person |
None of these is returned by the API after it is saved. Where the API reports on one, it says only whether a value is stored (for example hasClientSecret), or shows a secret once at creation (an outbound webhook's signing secret).
A test account for live testing stays with its pentest, encrypted, so that a retest can sign in again. It is deleted when the pentest is deleted by retention or erasure.
#How they are encrypted
| Aspect | As implemented |
|---|---|
| Cipher | AES-256-GCM, with a fresh random 96-bit nonce for every value and a full 128-bit authentication tag |
| Key per owner | The encryption key is not the service's key itself. For each organisation, and for each person in the case of authenticator secrets, a separate key is derived from the service's key with HKDF-SHA256, with the owner's identity as input |
| Owner binding | The owner's identity is also authenticated as part of the value (GCM additional authenticated data) |
| Organisation salt | An organisation can have a random 32-byte salt that is mixed into its key. Replacing the salt makes everything encrypted for that organisation before then impossible to open |
What scoping by owner gives you: a value encrypted for one organisation does not open in a record of another. If a defect, or someone with write access to the database, copied one organisation's stored credential into another organisation's record, the service would refuse to open it, because both the key and the authenticated owner would be wrong.
An authenticator secret belongs to a person rather than an organisation, so it keeps working whichever organisations that person joins or leaves.
#Key rotation
The service's key can be rotated. The service holds a key ring: the current key, with which everything new is encrypted, and earlier keys, which are kept only to open what they encrypted until those values have been re-encrypted under the current key. Each stored value records which key it was encrypted with, so rotation does not interrupt your integrations.
#Erasing an organisation's secrets
When an owner erases data with the all scope, Nullgap deletes the records and then replaces the organisation's salt. From then on, a copy of one of the deleted encrypted values that still existed somewhere could not be opened, because the key that opened it can no longer be derived.
- The few secrets that erasure keeps, such as the single sign-on client secret, are re-encrypted under the new salt in the same step, so they keep working.
- If a pentest with a test account is still queued or running, the salt is left as it is and the audit entry says so. Erase again once it has finished.
- The salt is stored with the organisation's data. Replacing it therefore does not reach a backup of the service made earlier, which holds the earlier salt next to the earlier values. Erasure is complete once such backups have aged out.
See erase data.
#What this encryption does not do
- It is not a key that you hold. The service can open these values, as it must in order to call your integrations. Anyone who held both the service's key and a copy of the database could derive every organisation's key. Per-owner keys limit mix-ups, tampering and the value of a single leaked field; they do not put your secrets beyond the service's own reach.
- It covers the secrets listed above, not everything. Findings, code excerpts, run logs and the audit log are not encrypted field by field. Their protection at rest is the storage encryption of the hosting, which is covered by your agreement with Nullgap. The console says the same on the Data protection page: "Integration credentials: AES-256-GCM. Database: per your hosting agreement."
#Values stored only as hashes
These are never stored in a form that can be read back. Nullgap keeps a SHA-256 hash and compares hashes when the value is presented.
| Value | Notes |
|---|---|
| API keys | Shown once when created or rotated. Only the hash and the first characters (the prefix shown in the list) are kept |
| Session tokens | The browser holds the token; the service holds its hash. Someone reading the database cannot use what they find to act as a signed-in person |
| SCIM tokens | Shown once. Only the hash is kept |
| Invitation links | The token in the link is kept only as a hash, which is why a lost link cannot be shown again |
| Recovery codes | Shown once. Each is kept as a hash and removed when used |
| Single-use sign-in state | The values that tie a single sign-on round trip to one browser are kept as hashes and deleted on use |
Because of this, a lost API key, SCIM token or set of recovery codes cannot be recovered, only replaced.
Not secret, and stored as they are: passkey public keys (the private key never leaves your device or password manager) and your identity provider's SAML signing certificates.
#Encryption in transit
#To Nullgap
The console and the API are served over HTTPS. When they are:
- the session cookie is marked
SecureandHttpOnly, usesSameSite=Lax, and carries the__Host-prefix, which binds it to the exact host so that a sibling subdomain cannot set or overwrite it; - every request from the console that changes something must also carry a matching anti-forgery header, so another site cannot ride on your session;
- API keys travel only in the
Authorizationheader. Never put a key in a URL.
#From Nullgap to your systems
Addresses you give Nullgap to call must use HTTPS, and the product refuses the ones that do not:
| Destination | Rule |
|---|---|
| Outbound webhook endpoints | https only. Each delivery is also signed, so you can check it came from Nullgap and was not altered: see webhooks |
| OIDC issuer and its endpoints | https, on a public host, with no credentials in the address |
| SAML sign-in address | https. SAML responses are additionally signed by your identity provider and checked against the certificates you stored |
| Jira | A public https address |
Events coming in from code hosts, Jira and Slack are authenticated with the shared secret or signature of that connection before anything is done with them.
#Passwords
Nullgap stores no passwords, because it has none: people sign in with an e-mail link, their GitHub account, a passkey or your identity provider. See multi-factor sign-in and passkeys.
#Related
Nullgap documentation